as per http://dev.mysql.com/doc/refman/5.0/en/replication-options-binary-log.html --sync-binlog=0
is the default. However, they recommend using '1', which indicates commit 1x per statement (if autocommit is on), or 1 per transaction. The docs say that this is very slow unless you have battery backed cache. Considering how many people were hurt by the lack of sync and missing schema, also considering that technically HDFS-265 doesn't actually flush to DISK, just ensures all 3 datanodes have copies of the data, we should be faster than MySQL with this setting set to 1, but not the fastest we could ever be. -ryan On Sat, Nov 14, 2009 at 6:58 PM, Chad Walters <chad.walt...@microsoft.com> wrote: > As a possible point of reference: What is the default behavior of MySQL? > > Chad > > -----Original Message----- > From: Ryan Rawson [mailto:ryano...@gmail.com] > Sent: Saturday, November 14, 2009 5:34 PM > To: hbase-dev@hadoop.apache.org > Subject: Re: Should we change the default value of > hbase.regionserver.flushlogentries for 0.21? > > The obvious value is '1'. But then we will be getting emails asking why > HBase is so slow... > > But we should still ship with '1' i think. > > On Sat, Nov 14, 2009 at 5:08 PM, Dave Latham <lat...@davelink.net> wrote: >> I tend to think that by default, all edits should be synced. Once a >> HTable.put method returns, the client can count on that data not being >> lost. The client can then disableAutoFlush, adjust its write buffer >> and use commit when it doesn't need every individual write flushed. I >> am definitely curious to hear the thoughts of the developers and other users >> however. >> Just my 2 cents. >> >> Dave >> >> On Sat, Nov 14, 2009 at 4:37 PM, Jean-Daniel Cryans >> <jdcry...@apache.org>wrote: >> >>> Hi dev! >>> >>> Hadoop 0.21 now has a reliable append and flush feature and this >>> gives us the opportunity to review some assumptions. The current situation: >>> >>> - Every edit going to a catalog table is flushed so there's no data loss. >>> - The user tables edits are flushed every >>> hbase.regionserver.flushlogentries which by default is 100. >>> >>> Should we now set this value to 1 in order to have more durable but >>> slower inserts by default? Please speak up. >>> >>> Thx, >>> >>> J-D >>> >> > >
